The Transims microsimulation approach

When designing a traffic microsimulation model, the first idea might be to measure all aspects of human driving and put them in algorithmic form into the computer. Unfortunately, such attempts cause many problems. The first is a data collection problem, because one can certainly not measure ``all'' aspects of human driving and is thus faced with the double sided problem that the necessary data collection process is extremely costly and still selective. Second, what if the emergent flow properties of such a model are clearly wrong, for example producing an hourly flow rate that is much too high?

For that reason, the Transims (TRansportation ANalysis and SIMulation System (112)) microsimulation starts with a minimal approach. A minimal set of driving rules is used to simulate traffic, and this set of rules is only extended when it becomes clear that a certain important aspect of traffic flow behavior cannot be modeled with the current rule set.^32.2 Besides the conceptual clarity, this also has the advantage that it is usually computationally fast - minimal models have few rules and thus run fast on computers.

The last paragraph leaves open what the ``important aspects'' are. In our view, this can only be decided in the proper context, i.e. when the question or problem area of application is known. The questions that Transims is currently designed for are transportation planning questions. These questions have traditionally been approached using traffic assignment models based on link performance functions (link capacity functions). Link performance functions are known to be dynamically wrong in the congested regime (95); they simply do not model queue build-up when demand is higher than capacity.

The most important result of a transportation microsimulation in that context should be the delays, since they dominate travel times, and also hinder discharge of the transportation system, thus leading to grid-lock. Delays are caused by congestion, and congestion is caused by demand being higher than capacity. This implies that the first thing the Transims traffic microsimulation has to get right are capacity constraints (and possibly their variance). Capacity constraints are caused by a variety of effects:

• Undisturbed roadways such as freeways have capacity constraints given by the maximum of the flow-density diagram.

• Typical arterials have their capacity constraints given by traffic lights.

• In the case of unprotected turning movements (yield, stop, ramps, unprotected left, etc.), the capacity constraints are given as a function of opposing traffic flows. For example, the number of vehicles making an unprotected left turn depends on the oncoming traffic.

Building a simulation which can be adjusted against all these diagrams seems a hopeless task given the enormous amount of degrees of freedom. The Transims approach for that reason has been to generate the correct behavior from a few much more basic parameters. The correct behavior with respect to the above criteria can essentially be obtained by adjusting two parameters: (i) The value of a certain asymmetric noise parameter in the acceleration determines maximum flow on freeways and through traffic lights; (ii) the value of the gap acceptance determines flow for unprotected movements.

It needs to be emphasized again that these remarks are only valid in our context: There are many questions for which the models need to have a higher fidelity, and then more details, higher resolution, etc. may need to be added (e.g. (125,118)).

There is sometimes debate whether the model we thus obtain is truly ``microscopic''. We use the term ``microscopic'' with respect to the resolution of the model, i.e. a model is microscopic as soon as it allows the identification of individual particles (here cars). The proposed area of application, though, is where traditionally more macroscopic models have been used (29,95,107,54).